chocodata-mcp

MCP server for the Chocodata web scraping API. Gives Claude, Cursor, Cline, Continue and any other MCP-compatible client tools to scrape the web, so an agent can pull live product prices, search results and social profiles without you writing glue code.

Runs over stdio. Node 18+.

Setup

Add it to your client's MCP config. It runs on demand via npx, so there is no global install and no version drift:

{
  "mcpServers": {
    "chocodata": {
      "command": "npx",
      "args": ["-y", "chocodata-mcp"],
      "env": { "CHOCODATA_API_KEY": "asa_live_YOUR_KEY" }
    }
  }
}

Get a key at chocodata.com. The free tier is 1,000 requests, one time, no card.

Claude Desktop: claude_desktop_config.json. Cursor: Settings, then MCP, then Add. Restart the client after editing, then ask it something like "what does this Amazon ASIN cost in Germany?"

Tools

Parameters follow the API's real contract rather than a uniform invention: amazon_product takes query, walmart_product takes url or id, and the Bing tools take q. Each tool's schema documents its own parameters, so the agent picks the right ones.

The generic tool

scrape reaches everything else in the catalog:

target:   "reddit"
resource: "post"
params:   { "post_id": "627akk", "subreddit": "askscience" }

Browse every target and its parameters at chocodata.com/scraper-api.

Errors

Failures come back as readable tool errors rather than stack traces, and say whether retrying will help:

Chocodata error 404 item_not_found: That TikTok handle does not resolve:
it has been deleted, renamed, or is unavailable in this region.
Request id: d2f460a7-...

You are only charged on a 2xx. Errors cost no credits. Retryable failures (429, 5xx, network) are retried automatically with backoff before the agent ever sees them.

Rate limits

120 requests per key per 60 seconds, plus a per-plan concurrency ceiling. An agent looping over many items can hit that; the server backs off and retries rather than failing the whole tool call.

License

MIT